Developing morphological computation in tensegrity robots for controllable actuation

Abstract

Conventionally control can be achieved by attempting to simplify complex dynamics. The field of morphological computation explores how mechanical complexity can be advantageous. In this paper we demonstrate morphological computation in tensegrity robots. We present a novel approach to tensegrity actuation and explore the capabilities of our self-evolving system. Methods of finding desirable gaits through both hand selection and evolution are described and the effectiveness of the system is demonstrated by our robot's ability to pursue a moving target. We conclude with a discussion of a bootstrapped system with the potential of significantly reducing evolution time and need for user presence.